Predictive Maintenance of Railway Point Machines

From the desk of

Sandeep Patalay

Predictive Maintenance of Railway
Point Machines

Sandeep Patalay

Senior Systems Engineer, CMC Ltd

sandeep.patalay@gmail.com

Abstract— The railway points (switches) are vital component of any Railway Interlocking system. Regular maintenance of points is required to keep them in operating condition. Present maintenance of points involves frequent inspection by maintenance staff and is not fool proof. Currently Electronic Monitoring systems are available which only logs the event and does not give any predictive analysis about the health of the points subsystem. This paper discusses a new approach for maintenance and diagnosis of railway points which is capable of remote monitoring and is intelligent enough to give predictive maintenance reports about the railway point’s health. This reduces the effort and huge costs in reducing manual monitoring and also it fool proof avoiding accidents. Distributed data gathering and centralized data processing methods have been discussed that not only report the fault but also give predictive measures to be taken by the field staff to avoid catastrophic failures.

Introduction

Railways traverse through the length and breadth of our country covering 63,140 route kms, comprising broad gauge (45,099 kms), meter gauge (14,776 kms) and narrow gauge (3,265 kms). The most important part of the railways to carry out operations like safe movement of trains and communications between different entities is Signalling. The Railway signalling is governed by a concept called Interlocking. The main component of the interlocking is the Railways Points consisting of DC electrical motors to switch the rails to a different route.  These vast and widespread assets to meet the growing traffic needs of developing economy is no easy task and makes Indian Railways a complex cybernetic system. The current mechanism in place to maintain the railway points are completely manual and requires large pool of maintainers to check the validity of the point machine and the related point infrastructure regularly, this process employed is neither cost effective nor fool proof. By employing the traditional method of manual maintenance, the rail operators do not have any prior warning for replacement or repair of points. The discussion in this paper mainly focuses on development of a system that not only monitors the points remotely without manual intervention, but also diagnosis the problem in the point thus saving human lives and huge manual maintenance costs. The motivation for developing a predictive maintenance system for Railway Points is as follows:

1. To use an array of sensors to monitor all relevant parameters, in order to provide advanced warning of degradation prior to railways points failure.
2. To provide predictive maintenance reports about the point machines to the maintainers.
3. To provide continuous monitoring at both local and centralized locations.
4. To provide an automated archival record from which broad trends can be extracted from the entire railway asset base.
5. To provide, in the event of a catastrophic failure, the immediate past history to identify the cause.

Railway Points Structure

The following Figure 1 describes the architecture of railway points in operation.

Figure 1

Points, or switches as they are known, allow a rail vehicle to move from one set of rails to another. They are a ‘digital output device’ in that there are only two acceptable states for the point to be set in, ‘normal’, and ‘reverse’. Movement is carried out by way of a geared motor, which actuates the stretcher bar. Location or state detection is made by a two-position, polarized, magnetic stick contactor. A signal is fed back from these switches to the signal box where all point directions are controlled and monitored. The snap-action switches at the end of the stroke stop the machine and help brake the motor to help reduce any impact at the end of the travel. Two stretcher bars (Figure 1) make sure that the switch rails remain the correct distance apart – this can vary between installations depending on the curvature of the main rails, and the speed limit of that section of the track. There are usually two stretcher bars for each point machine. Any fault in this mechanism like poorly securing of the bolts holding these stretcher bars, loose bolts etc. may lead to deadly accidents.

Proposed Predictive Maintenance System Architecture

The proposed architecture of Predictive Maintenance System (PMS) for Railways points is discussed below using the Figure 2

Figure 2 Architecture of PMS

Sensors are used to measure Voltage, Current, load and temperature of Point Motor. The Throwing load sensor is used to measure the stress in the operating rod of the point machine. The sensor values are read on real time basis by the wayside device and sent to a central location for analysis. The wayside device uses GSM/GRPS network to transmit this data to a central location. The Central Station analyzes the data in real time and makes predictions on the point machines and stores them in to a database. The status of any point machine can be viewed using any internet browser in the central station. The Local station maintainers can view the data by logging in to the web server using any internet browser. Based on the Current consumption, the load sensor values and the point motor temperature, predictions are made for the maintenance or replacement of the Point Motors. The central location is a Web server based architecture, where anyone with a Web browser can login and see the details.

Data processing and analytics

The system has a database of current and load characteristics of good working railway points. This data is used as a reference for processing real time data received from the wayside units. The following figures show the current (i) and load sensor values plotted against time during point machine operation.

Figure 3 Current Characteristics

Figure 4 Load Force Characteristics

Data Processing Techniques

Various Signal Processing Techniques are available for analysis of real time data described below:

1)       Data Cluster method – This involves recording the characteristics of a parameter of a subsystem under different simulated conditions and then using this as a reference to validate the real time data. This method is different from template matching, since it not entirely based on matching the plotted characteristics.

2)       Template matching – Entails comparing complete data sets with pre-recorded examples of data resulting from known fault conditions. The method can be used effectively in some circumstances, provided a representation of the data that produces good discrimination between pattern classes can be made. However, this requires a substantial amount of experimentation with different transformations of the data sets to find such distinctions, and would be a computationally intensive process.

3)       Statistical and decision theoretic methods – Matches are made based on statistical features of the signal. For example, the mean and peak-to-peak value are evaluated for each vector, and plotted in feature space, whereby different patterns are distinguishable because they form clusters for each class that are located apart from the fully functioning case.

4)       Structural or syntactic methods – Involves deconstructing a pattern or vector into structural components, to enable comparisons to be made on more simple, sub-segments of data rather than a complete vector. Mathematically, these methods are similar to fractal-based compression routines.

The method that was of specific interest to this project was to use a data clustering methodology where a database of good measurements as well as load sensor data readings under various simulated faults in the laboratory on some specimen railway points is stored  and then the real time load sensor data is plotted against it. This generates very unique clusters of data points which represent each type of fault.

By applying the above techniques, we get clusters of fault data. We have found that these data clusters are unique in the sense that these represent different types of faults.

Figure 5 Force Data Clusters

          Types of faults detectable

1. Tight lock on reverse side (sand on bearers both sides) – Refers to the lock which holds the point in position after it has changed direction. This lock prevents the point from moving out of position because of vibration.
2. A 12-mm obstruction at toe on normal side – Simulates a piece of ballast impeding point motion between the toe of the switch rail (the mobile section of rail), and the stock rail.
3. Back drive slackened off at toe end on LHS – The drive to the midpoint of the switch rail is only loosely connected to the stretcher bar. The stretcher bar holds the mobile rails a fixed distance apart.
4. Back drive slackened off at toe end on RHS – Similar to the above.
5. Back drive tightened at heel end on RHS – Similar to the above.
6. Back drive tightened at heel end on LHS – Similar to the above.
7. Diode snubbing block disconnected – An electrical fault.
8. Drive rod stretcher bar loose on RHS – Connecting bar between the switch rails is loose. A dangerous fault.
9. Operational contact slackened off by four holes – Applies to the contact for detecting when the point has completed motion.